BJSS Helps University of Bath Track Room Occupancy with 90% Accuracy on AWS

The University of Bath has an ambitious data project to empower decision making by giving the right people the right data at the right time. It began exploring business use cases around data use in research, education, and space management using AWS. As part of the project, it wanted to collect solid, accurate data on exactly how many students attend lectures and seminars. It also wanted to gauge library and study space use. Better data would improve decisions about new buildings and enable better use of existing facilities. It could even provide a near real-time data source for students looking for quiet spaces in the library or elsewhere.

One issue with utilization of space was comparing timetable data with actual occupancy. Previously, this relied on static timetable data and manual observations of room use, coupled with data on the numbers of people entering a building. But these manual processes were time-consuming. Furthermore, the results were historical, revealing when and how a space had been used, not its current state. So the university investigated the use of different kinds of sensor data. The university had existing sensors in the form of break beam detectors at the entrance points of some teaching areas, along with data from Wi-Fi routers. It also looked at CO2 monitors. However, it didn’t know how reliable or accurate this data was and how usable it would be.

The University of Bath chose to work with BJSS after a recommendation from AWS. “We really appreciate BJSS’s agility and flexibility,” says Gavin Edwards, chief data and technology officer at the University of Bath. “This was a blue-sky project without a clear roadmap—they really supported this strategy.”

Initial work compared Wi-Fi data with data from break beams placed at room doorways. When it added sensor data from other sources and real-world observations, the team realized that Wi-Fi data, with some cleaning and processing, could provide accurate results on its own. Errors could occur, for example, when students gathered outside a room or lecture theater at changeover time. The team also manually mapped the Wi-Fi coverage within the rooms to discover potential blind spots or crossover areas where devices might register on two routers. Although there was a margin of error of 10–15 percent, this is well within the parameters used in the higher education sector of quartiles of room occupancy—empty, 25 percent, 50 percent, 75 percent, or full.

The solution is easily scalable using a variety of AWS services. Data came in a variety of formats and had to be anonymized and ingested into Prometheus, an online time series database, before being sent to AWS using a virtual private network. The data is then ingested into AWS Fargate, which routes traffic into a three-step process. Raw data was first sent to an Amazon Simple Storage Service (Amazon S3) bucket labeled bronze. This data was then cleaned up using AWS Glue and Amazon Athena and moved into a silver bucket. Finally, after more AWS Glue and Amazon Athena processing and structuring, it was put into the gold bucket of curated and useful data.

The final stage of the solution shows the data, in near real-time, in a visual and easy-to-understand format for staff and students. Dashboards were built using Time Series Grafana, although many other platforms, including PowerBI, could have been used. The dashboards can display the data on a floorplan or as an animation showing percentage of occupancy for each room.